Commits

Branch selector

 main

User selector

yifuwang

Datepicker

All time

Commit History

Commits on Nov 26, 2024

• [IntraNodeComm] fix a recent breakage (pytorch#141200)

  Show description for 6a22cae
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  6a22cae

  Copy full SHA for 6a22cae

Commits on Nov 18, 2024

• Fix unintended deprecation warning in torch.distributed.optim (pytorch#140889)

  Show description for 3d26c08
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  3d26c08

  Copy full SHA for 3d26c08

Commits on Nov 17, 2024

• [SymmetricMemory] introduce user-facing APIs empty() and rendezvous() (pytorch#139677)

  Show description for 5a7e147
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  5a7e147

  Copy full SHA for 5a7e147

• [SymmetricMemory] support specifying group_name at rendezvous time (pytorch#139529)

  Show description for ab5c885
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  ab5c885

  Copy full SHA for ab5c885

Commits on Nov 14, 2024

• [SymmetricMemory] fix a bug in symm_mem::memset32_ where the ops fails when offset=0 (pytorch#140129)

  Show description for 02d0c43
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  02d0c43

  Copy full SHA for 02d0c43

• [SymmetricMemory] fix a bug where get_signal_pad() returns a tensor backed by a buffer ptr instead of a signal_pad ptr (pytorch#140128)

  Show description for 684db9b
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  684db9b

  Copy full SHA for 684db9b

• [SymmetricMemory] allow overlapping devices for testing (pytorch#140127)

  Show description for c3d61bd
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  c3d61bd

  Copy full SHA for c3d61bd

Commits on Nov 11, 2024

• [SymmetricMemory] improve the API for stream_write_value32 (pytorch#139934)

  Show description for 0a0915f
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  0a0915f

  Copy full SHA for 0a0915f

• [SymmetricMemory] improve the API for stream_write_value32 (pytorch#139934)

  Show description for 2f3a5a1
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  2f3a5a1

  Copy full SHA for 2f3a5a1

Commits on Nov 8, 2024

• [experimental] async-tp impl with cutlass-based, progress aware kernel (pytorch#139227)

  Show description for 1659e24
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  1659e24

  Copy full SHA for 1659e24

Commits on Nov 7, 2024

• [experimental] async-tp impl with cutlass-based, progress aware kernel (pytorch#139227)

  Show description for 5203138
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  5203138

  Copy full SHA for 5203138

Commits on Nov 5, 2024

• [SymmetricMemory] introduce a binding for cuMemset32Async (pytorch#138755)

  Show description for ee42a99
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  ee42a99

  Copy full SHA for ee42a99

Commits on Nov 3, 2024

• [SymmetricMemory] introduce a binding for cuMemset32Async (pytorch#138755)

  Show description for 924e726
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  924e726

  Copy full SHA for 924e726

Commits on Nov 1, 2024

• [SymmetricMemory] expose signal_pads as tensors in Python (pytorch#138754)

  Show description for 0dbc284
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  0dbc284

  Copy full SHA for 0dbc284

• [Inductor] fix a compilation time regression caused by user-visible output handling (pytorch#139420)

  Show description for e6e140c
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  e6e140c

  Copy full SHA for e6e140c

Commits on Oct 30, 2024

• Preliminary registered-buffer collective support via Inductor (pytorch#138029)

  Show description for 7765d1e
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  7765d1e

  Copy full SHA for 7765d1e

• get_symm_mem_workspace(): print helpful error during graph capture (pytorch#138028)

  Show description for 421473c
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  421473c

  Copy full SHA for 421473c

Commits on Oct 26, 2024

• [Inductor] improve the stride preservation logic of user-visible outputs (pytorch#136732)

  Show description for 3a6f014
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  3a6f014

  Copy full SHA for 3a6f014

Commits on Oct 25, 2024

• Removed the typing information for already deleted ProcessGroupCudaP2P (pytorch#138753)

  Show description for ed9169d
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  ed9169d

  Copy full SHA for ed9169d

Commits on Oct 17, 2024

• [SymmetricMemory] fix a race condition in _pipelined_produce_and_all2all that can cause correctness issues for very small chunk_producers (pytorch#138126)

  Show description for c69f451
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  c69f451

  Copy full SHA for c69f451

Commits on Oct 16, 2024

• [SymmetricMemory] fix incorrect numel caculations that are using int as std::accumulate's accumulator (pytorch#138038)

  Show description for 80f3ee4
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  80f3ee4

  Copy full SHA for 80f3ee4

Commits on Oct 15, 2024

• Improve the scheduling of _pipelined_multi_all_gather_and_consume (pytorch#137850)

  Show description for 5d5783a
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  5d5783a

  Copy full SHA for 5d5783a

• Improve the scheduling of _pipelined_produce_and_all2all (pytorch#137836)

  Show description for 2ae1a4c
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  2ae1a4c

  Copy full SHA for 2ae1a4c

• [fused_all_gather_scaled_matmul] support rowwise scaling (pytorch#137805)

  Show description for ef541c1
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  ef541c1

  Copy full SHA for ef541c1

• [fused_scaled_matmul_reduce_scatter] support rowwise scaling (pytorch#137738)

  Show description for 05edaea
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  05edaea

  Copy full SHA for 05edaea

• [SymmetricMemory] implement timeout for barrier(), put_signal() and wait_signal() (pytorch#137643)

  Show description for 91bc9dc
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  91bc9dc

  Copy full SHA for 91bc9dc

Commits on Oct 10, 2024

• [SymmetricMemory] set the storage_offset of tensors returned by get_buffer() to 0 (pytorch#137569)

  Show description for ea83c78
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  ea83c78

  Copy full SHA for ea83c78

Commits on Oct 9, 2024

• [CudaDMAConnectivityDetector] improve the detection robustness (pytorch#137530)

  Show description for aafa025
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  aafa025

  Copy full SHA for aafa025

• [SymmetricMemoryOps] use float32 as the accumulator type when accumulating bfloat16 with multimem.ld_reduce (pytorch#137529)

  Show description for fbaf9b6
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  fbaf9b6

  Copy full SHA for fbaf9b6

• [IntraNodeComm] replace all-reduce kernels with corresponding symm_mem ops (pytorch#137475)

  Show description for 39c5122
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  39c5122

  Copy full SHA for 39c5122

• [SymmetricMemoryOps] create an out-variant for multimem_one_shot_all_reduce (pytorch#137474)

  Show description for e6edfe3
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  e6edfe3

  Copy full SHA for e6edfe3

• [SymmetricMemory] fix a bug where numel calculation overflows when the tensor size is large (pytorch#137567)

  Show description for 38027b9
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  38027b9

  Copy full SHA for 38027b9

• [SymmetricMemoryOps] implement two-shot all-reduce (pytorch#137473)

  Show description for d3edf4e
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  d3edf4e

  Copy full SHA for d3edf4e

• [SymmetricMemoryOps] implement one_shot_all_reduce (pytorch#137472)

  Show description for 82e55b6
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  82e55b6

  Copy full SHA for 82e55b6

• [SymmetricMemoryOps] refine cross-device barriers (pytorch#137471)

  Show description for 5d83ee3
  

  yifuwang

  authored and

  pytorchmergebot

  committed
  5d83ee3

  Copy full SHA for 5d83ee3

Pagination

Previous

Next